The latest release of microsoft/agensh is v1.0. The provided information contains no changelog for this release, so no differences from the previous version can be confirmed. This repository is a multi-agent execution framework in which multiple agents, without a central coordinator, find and claim sub-tasks, share their results, and validate and merge them asynchronously. It is a Python project published by Microsoft and uses the MIT license.
This structure connects a workspace, a message interface, and a shared context so that agents can divide tasks and exchange results. The README explains that the way the shared context is composed includes elements taken from Gitea, Mattermost, and DeLM. It states that it scales to as many as 1,024 agents, and it also provides experimental run scripts and reproduction guidance. A structure that does not centrally coordinate the work of multiple cooperating agents is the core of this project.
That said, the figure of up to 1,024 agents is the scalability range the project states, not a measured result that allows judging performance or quality in actual work from the provided information alone. The experiments and reproduction guidance can serve as grounds for examining the structure, but no specific changes in v1.0 or performance comparisons were given. What can be confirmed at this point is the design direction and the execution guidance, while the real effect of adoption is at a stage that requires separate verification.